Demand Response Can Improve the Emission Benefits of Windidei.fr › sites › default › files ›...

Demand Response Can Improve the EmissionBenefits of Wind

Seyed Hossein Madaeni1 Ramteen Sioshansi2

1Pacific Gas and Electric Company

2The Ohio State University

Conference on The Economics of Energy MarketsToulouse, France

17–18 January, 2013

Madaeni and Sioshansi (PG&E and OSU) DR Can Improve Emissions Benefits of Wind 17–18 January, 2013 1 / 36

Disclaimer

The opinions expressed and conclusions reached are solely those ofthe authors and do not represent the official position of Pacific Gasand Electric Company.


Background

Renewable Growth

Tremendous growth in renewable deployment in the past decadeMostly wind (e.g., 238 GW wind and 67 GW of PV by end of 2011[IEA13])Many motivations:

Sustainable energy supplyMore cost competitive with alternativesSupply diversificationClimate changeOther environmental and emissions impacts of conventional alternatives


Background

Those Emissions

Renewables displace fossil fueled-generation and their emissionsNet effect is more nuanced, due to resource variability and uncertainty

Figure: Tehachapi Wind Generation in April 2005Source: California ISO

Need some mixture ofsupply- anddemand-side flexibilityto operate systemreliablyCurrently usesupply-side solutions(i.e., partially loadgenerators to provideramping and reserves)


Background

Why Renewable Integration Matters

0 50 100 150 200 2500

10

20

30

40

50

60

70

80

90

100

Generator Output [MWh]

Hea

t rat

e [G

J/M

Wh]

Figure: Measured Heat Rate of WolfHollow I CCGT Source: US EPA CEMs

0 50 100 150 200 2500

0.02

0.04

0.06

0.08

0.1

0.12

Generator Output [MWh]E

mis

sion

s R

ate

[kg/

GJ]

Figure: Measured NOx Rate of WolfHollow I CCGT Source: US EPA CEMs

You can get hit with double-whammy of higher heat rates and lesseffective emissions controls when units are partially loaded


Background

Net Emissions Gain?

Katzenstein and Apt [KA09] study these interactionsAssume an NG unit follows output of a single wind plant to balance itsreal-time outputUse five-minute wind data to simulate NG dispatchCombine with CEMs data to estimate CO2 and NOx emissions

CO2 emissions go down, but by less than back of the envelope‘displacement’ analysis predictsNOx emissions sensitive to type of controls—net increase if NG plant usesdry control technology


Background

What We Do

Use a case study based on ERCOT (Texas) system to:1 Revisit wind emissions analysis to determine what happens with a full wind

and conventional portfolio and real system operating rules2 Examine effects of using supply-side measures (RTP) to accommodate wind

variability and uncertainty and emissions impacts

We find:1 Wind decreases system-wide conventional generator efficiency and

increases marginal emissions rates (partial loading effect)2 Net CO2, SO2, and NOx reductions, since displacement outweighs efficiency

impacts3 RTP increases emissions due to rebound effect and changed diurnal load

profiles, but decreases costs associated with wind uncertainty and variability4 Taking these two effects together, RTP delivers better bang for the buck

(more emissions abatement per dollar of wind-integration cost)


Modeling Approach

Two Effects of Wind

Short-run analysis, with fixed conventional generation mix

Two short-run effects that we focus on:1 Cost

Although zero marginal cost, wind imposes external costsHave to commit excess capacity and dispatch it in real-time to balance variableoutput

2 EmissionsDisplacement of fossil-fueled generatorsEfficiency effects of partial loading


Modeling Approach

Model Overview

Model operations with unit commitment and economic dispatch modelsStochastic unit commitment model determines which generators tostartup and run day-ahead, accounting for future wind uncertaintyDeterministic real-time model determines generator dispatch based onactual wind‘Fast start’ units can be committed in real-time, if neededCases with RTP model demand response as a dispatchable resource,determine how much load to serve and how to serve it to maximize socialwelfare (based on assumed inverse demand function)All models operate at hourly timesteps and optimize operations in arolling fashion


Modeling Approach

Day-Ahead Stochastic Unit Commitment

Day-ahead model has atwo-stage scenario tree structureScenarios represent possiblewind realizationsStochastic model commits a mixof units that can feasibly serveload under a range of possiblewind scenarios (ξ ∈ Ξ)Generator commitments (ug,t ,sg,t , and hg,t ) determined in firststage and arescenario-independent

root node

ξ3

ξ4

ξ2

ξ1

stage 2stage 1qg,t,ξ, ρ

spg,t,ξ, ρ

nsg,t,ξ,ug,t , sg,t ,hg,t

wt,ω,ξ, lt,ξ

Figure: Scenario Tree Schematic


Modeling Approach

Day-Ahead Stochastic Unit Commitment

Generator dispatch and reserves(qg,t,ξ, ρ

spg,t,ξ, and ρns

g,t,ξ), windgeneration (wt,ω,ξ), and loadserved (lt,ξ) determined insecond stage and arescenario-dependentObjective function maximizesexpected social welfare (integralof inverse demand function, lessgeneration cost)Includes load-balance, spinningand non-spinning reserve,ramping, minimum up- anddown-time constraints

Model Formulation

root node

ξ3

ξ4

ξ2

ξ1

stage 2stage 1qg,t,ξ, ρ

spg,t,ξ, ρ

nsg,t,ξ,ug,t , sg,t ,hg,t

wt,ω,ξ, lt,ξ

Figure: Scenario Tree Schematic


Modeling Approach

Real-Time Economic Dispatch

Deterministic model, has same model structure (i.e., maximize expectedsocial welfare subject to same constraint types)Use actual wind availability

Generally different than any of the scenarios modeled day-ahead

Generator commitments fixed based on day-ahead solution‘Fast start’ units can be started up, if neededDetermines actual generator dispatch and load served, based on actualwind and day-ahead commitment


Modeling Approach

Emissions Estimation

Translate generator commitmentand dispatch into fuel use, basedon heat ratesUse input-based emissions rateestimatesConstant CO2 rates, since theseare uncontrolledNon-parametric estimates of SO2and NOx rates, based on CEMsdata

Non-Parametric Emissions Estimate

0 500 1000 1500 2000 2500 30000

0.02

0.04

0.06

0.08

0.1

0.12

Fuel burned [GJ]In

put−

base

d em

issi

ons

rate

[kg/

GJ]

Actual CEMs dataNon−parametric rate estimate

Figure: Non-parametric Estimate of NOx

Rate of Wolf Hollow I CCGT Source: USEPA CEMs


Modeling Approach

Case Study

Based on the ERCOT system, used in previous wind-integration studies[SS09, Sio10, MS13a, MS13b]Use 2005 conventional generator and load dataModel a high-wind case with 14 GW of wind (14% of generating capacity)

Table: Conventional Generation Mix [%]

Type Capacity Base Case Generation

Natural Gas 75 41Coal 19 45Nuclear 6 14


Modeling Approach

Case StudyDemand Response

Model demand response as a dispatchable resource the system operatorcan use to balance the systemUse an inverse demand function to represent willingness to pay forenergyCalibrate demand function in each hour by assuming an own-priceelasticity (cross-price elasticities assumed to be zero) and fixing:

pt (lt ) = pret ,

where:lt is historical demandpret is historical retail electricity electricity price

Load fixed and equal to lt in fixed-load cases


Modeling Approach

Case StudyWind Modeling

14 GW of wind based on sites built through end of 2011Modeled actual wind availability (used in real-time dispatch model) fromNREL’s Western Wind Resources Dataset (WWRD)

Wind generators are associated with locations in WWRDActual wind modeled as:

Ωω · φω,t ,

where:Ωω is nameplate capacityφω,t ∈ [0, 1] is fraction available in hour t


Modeling Approach

Case StudyWind Modeling

Day-ahead wind forecasts aregenerated by adding a forecasterror term to actual wind availability:

φω,t + εt ,

Assume a serially autocorrelatederror structure [LH07]:

εt = νt + ζ · εt−1

where νt is hour-t innovation, withunbiased truncated GaussiandistributionRandomly generate 1000 samplepaths and use SCENRED to reduceto four scenarios [DGKR03]

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 240

0.5

1

1.5

2

2.5

3

3.5

4

Hour

Act

ual/F

orec

aste

d W

ind

Ava

ilabl

e [G

W]

ActualScen. 1Scen. 2Scen. 3Scen. 4

Figure: Modeled Day-Ahead WindForecasts and Actual Wind


Modeling Approach

Cases Modeled

Compare system costs and emissions in five cases to determine effectsof windCases with ‘perfect foresight’ assume wind availability is knownday-ahead, no forecast or scenario tree used

Table: Cases Modeled

Case Wind DA Wind Loads

Base None n/a FixedWind Fore High Forecast FixedWind PF High Perfect Foresight FixedDR Fore High Forecast Price-ResponsiveDR PF High Perfect Foresight Price-Responsive


Results

Emissions Effects of WindFixed Load Case

Emissions in Base and Wind Fore cases≈43 TWh of conventional generation (27% coal) displaced by windEmissions reductions, increasing in forecast error variance

Table: Annual Generation and Generator Emissions

Wind Forecast Generation [TWh] EmissionsError Variance Coal NG CO2 [Mt] SO2 [kt] NOx [kt]

No Wind 134.8 122.8 198.5 451.2 140.40.0025 122.9 91.0 172.1 400.7 124.20.0100 122.7 91.1 171.9 399.7 124.00.0225 122.5 91.4 171.8 398.8 123.8


Results

Emissions Effects of WindFixed Load Case

Higher variance means less accurate forecastsDay-ahead wind scenarios are more variable, requiring greater rampingCommit more NG units (in place of coal), shifting load to themNG units are more lightly loaded, giving lower heat rates and less efficientSO2 and NOX control

Table: Average Natural Gas Plant Loading and Emissions Rates

Wind Forecast Plant Emissions Rates [g/MWh]Error Variance Loading [%] SO2 NOx

0.0025 34.1 3.59 349.50.0100 30.2 3.59 350.30.0225 28.1 3.61 350.6


Results

Emissions Effects of WindNOx Decreases

NOx decreases, despite wind variability and uncertainty (contraKatzenstein and Apt)Observe same loss of generator and emissions control efficiencyTwo important distinctions with their assumptions:

They implicitly assume 100% reserves for each wind plant (i.e., model adedicated NG plant following wind output); we use stochastic model to‘dynamically’ determine reserve needs, exploit spatial smoothing of windavailabilityDon’t capture effect of wind on mix of committed generators (i.e., shift awayfrom coal to NG, due to ramping needs)


Results

Emissions Effects of WindPrice-Responsive Load Case

Table: Annual Emissions Increase (Between DR Fore and Wind Fore Cases) [%]

Wind Forecast Demand Elasticity -0.1 Demand Elasticity -0.3Error Variance CO2 SO2 NOx CO2 SO2 NOx

0.0025 0.9 3.0 0.5 2.3 7.3 0.70.0100 0.9 3.1 0.6 2.3 7.5 0.80.0225 0.9 3.1 0.6 2.3 7.6 0.9

Emissions increase when RTP is introducedPrimarily due to a change in the diurnal load pattern—off-peak energy isrelatively cheap, loads shift toward those hoursCoal generation is marginal during off-peak hours (coal increases from57% to 60% of thermal generation)Also have a rebound effect—wind suppresses real-time prices, givingoverall demand increase


Results

Cost Effects of Wind

Wind has an ancillary cost effect, due to variability and uncertaintyWe measure this ‘wind-uncertainty cost’ as difference between operatingcost with wind forecasts to counterfactual case with perfect foresight[DGMS05, SMDP07, DJK+07]In fixed load case, this is welfare loss between Wind PF and Wind ForecasesWith price-responsive loads, compare DR PF and DR Fore cases

Table: Wind-Uncertainty Cost [$/MWh of Wind]

Wind Forecast Demand ElasticityError Variance Fixed Load −0.1 −0.3

0.0025 1.81 0.25 0.020.0100 3.79 0.99 0.020.0225 6.11 1.89 0.04


Results

The ‘Bottom Line’ of Wind

Wind has two primary ancillary effects on the power system:1 External uncertainty cost2 Emissions abatement

Capture the two by computing emissions averted per dollar ofwind-uncertainty cost incurredAlthough RTP errodes some emissions benefits of wind, the lowerwind-uncertainty cost more than makes up for thisLess accurate wind forecasting is a costly means of reducing emissions

Table: Annual Emissions Averted Per Dollar of Wind-Uncertainty Cost

Wind Forecast Without RTP With RTPError Variance CO2 SO2 NOx CO2 SO2 NOx

[t/$] [kg/$] [kg/$] [t/$] [kg/$] [kg/$]

0.0025 3 6 0 23–232 35–216 1–100.0100 2 3 0 6–228 9–215 0–110.0225 1 2 0 3–125 5–119 0–6


Conclusions

To Summarize

We explore interactions between wind and demand response in energysystems, with focus on emissionsRevisit some counterintuitive findings regarding net emissions impacts ofwindShow that when the total power system ‘portfolio’ and operating practicesare accounted for, wind delivers overall emissions reductionsDemand response is a cost- and emissions-effective way ofaccommodating wind variability and uncertaintyReduced wind-integration costs due to demand response may also lowertechnical and financial barriers to entryDemand response can also reduce wind curtailment, when that becomesan issue [LCR03, DM07, SS09, TO11]


Conclusions

Caveats and Future Work

Short-run analysis only, not much detailed work examining long-terminvestment and issues or benefits from demand responseUnclear how other renewables (e.g., solar) may benefit from demandresponseDifferences possible due to diurnal and seasonal solar patterns being sodifferent to wind


Conclusions

Questions?


Conclusions

References I

Jitka Dupacová, Nicole Gröwe-Kuska, and Werner Römisch.Scenario reduction in stochastic programming.Mathematical Programming, 95:493–511, March 2003.

Edgar A. DeMeo, William Grant, Michael R. Milligan, and Matthew J.Schuerger.Wind plant integration.IEEE Power and Energy Magazine, 3:38–46, November-December 2005.

Edgar A. DeMeo, Gary A. Jordan, Clint Kalich, Jack King, Michael R.Milligan, Cliff Murley, Brett Oakleaf, and Matthew J. Schuerger.Accommodating wind’s natural behavior.IEEE Power and Energy Magazine, 5:59–67, November-December 2007.

Paul Denholm and Robert M. Margolis.Evaluating the limit of solar photovoltaics (PV) in traditional electric powersystems.Energy Policy, 35:2852–2861, 2007.


Conclusions

References II

IEA.Renewable Energy.online athttp://www.iea.org/aboutus/faqs/renewableenergy/, 2013.accessed on 11 January, 2013.

Warren Katzenstein and Jay Apt.Air emissions due to wind and solar power.Environmental Science and Technology, 43:253–258, January 2009.

Lower Colorado River Authority, Austin, Texas.Study of Electric Transmission in Conjunction with Energy StorageTechnology, August 2003.Prepared for Texas State Energy Conservation Office.


http://www.iea.org/aboutus/faqs/renewableenergy/

Conclusions

References III

Clyde Loutan and David Hawkins.Integration of renewable resources.Technical report, California Independent System Operator, November2007.

Seyed Hossein Madaeni and Ramteen Sioshansi.The impacts of stochastic programming and demand response on windintegration.Energy Systems, 2013.forthcoming.

Seyed Hossein Madaeni and Ramteen Sioshansi.Using demand response to improve the emission benefits of wind.IEEE Transactions on Power Systems, 2013.forthcoming.


Conclusions

References IV

Ramteen Sioshansi.Evaluating the impacts of real-time pricing on the cost and value of windgeneration.IEEE Transactions on Power Systems, 25:741–748, April 2010.

J. Charles Smith, Michael R. Milligan, Edgar A. DeMeo, and BrianParsons.Utility wind integration and operating impact state of the art.IEEE Transactions on Power Systems, 22:900–908, August 2007.

Ramteen Sioshansi and Walter Short.Evaluating the impacts of real-time pricing on the usage of windgeneration.IEEE Transactions on Power Systems, 24:516–524, May 2009.

Aidan Tuohy and Mark O’Malley.Pumped storage in systems with very high wind penetration.Energy Policy, 39:1965–1974, April 2011.


Modeling Details

Modeling Details


Modeling Details

Day-Ahead Stochastic Unit CommitmentModel Formulation

max∑ξ∈Ξ

∑t∈T

πξ ·

∫ lt,ξ

0pt (x)dx −

∑g∈G

[cv

g (qg,t,ξ) + cng · ug,t + cs

g · sg,t] ;

s.t. lt,ξ = wt,ξ +∑g∈G

qg,t,ξ; ∀ t ∈ T , ξ ∈ Ξ;

∑g∈G

ρspg,t,ξ ≥ η

sp · lt,ξ; ∀ t ∈ T , ξ ∈ Ξ;

∑g∈G

(ρspg,t,ξ + ρns

g,t,ξ) ≥ (ηsp + ηns) · lt,ξ; ∀ t ∈ T , ξ ∈ Ξ;

K−g · ug,t ≤ qg,t,ξ; ∀ g ∈ G, t ∈ T , ξ ∈ Ξ;

qg,t,ξ + ρspg,t,ξ ≤ K +

g · ug,t ; ∀ g ∈ G, t ∈ T , ξ ∈ Ξ;


Modeling Details


qg,t,ξ + ρspg,t,ξ + ρns

g,t,ξ ≤ K +i ; ∀ g ∈ G, t ∈ T , ξ ∈ Ξ;

0 ≤ ρspg,t,ξ ≤ ρ

spg · ug,t ; ∀ g ∈ G, t ∈ T , ξ ∈ Ξ;

0 ≤ ρnsg,t,ξ ≤ ρns

g ; ∀ g ∈ G, t ∈ T , ξ ∈ Ξ;

R−g ≤ qg,t,ξ − qg,t−1,ξ; ∀ g ∈ G, t ∈ T , ξ ∈ Ξ;

qg,t,ξ − qg,t−1,ξ + ρspg,t,ξ + ρns

g,t,ξ ≤ R+g ; ∀ g ∈ G, t ∈ T , ξ ∈ Ξ;

t∑y=t−τ+

g

sg,y ≤ ug,t ; ∀ g ∈ G, t ∈ T ;

t∑y=t−τ−g

hg,y ≤ 1− ug,t ; ∀ g ∈ G, t ∈ T ;

sg,t − hg,t = ug,t − ug,t−1; ∀ g ∈ G, t ∈ T ;


Modeling Details


0 ≤ wt,ξ ≤ wt,ξ; ∀ t ∈ T , ξ ∈ Ξ;

lt,ξ ≥ 0 ∀ t ∈ T , ξ ∈ Ξ; andug,t , sg,t ,hg,t ∈ 0,1; ∀ g ∈ G, t ∈ T .

Model Overview


Modeling Details

Emissions EstimationNon-Parametric Estimate

Estimate emissions rate as afunction of fuel use based on USEPA CEMs dataEmissions rate of species pestimated as:

φg,p(f ) =

∑Nn=1 K

(f−f n

gh

)φn

g,p∑Nn=1 K

(f−f n

gh

) ,

where:φn

g,p and f ng are emissions rate

and fuel use in CEMs dataK (·) is a kernel function, taken tobe Gaussian densityh is bandwidth

Emissions Estimation

0 500 1000 1500 2000 2500 30000

0.02

0.04

0.06

0.08

0.1

0.12

Fuel burned [GJ]

Inpu

t−ba

sed

emis

sion

s ra

te [k

g/G

J]

Actual CEMs dataNon−parametric rate estimate

Figure: Non-parametric Estimate of NOx

Rate of Wolf Hollow I CCGT Source: USEPA CEMs


Demand Response Can Improve the Emission Benefits of Windidei.fr › sites › default › files ›...

Documents

Transcript of Demand Response Can Improve the Emission Benefits of Windidei.fr › sites › default › files ›...